Cluster States
Use the Cluster States option on the Actions menu of the Cluster Summary panel to change a cluster state or settings.
Select this option to put the cluster into a different state. Multiple options can be shown, depending on the current state of the cluster.
| If the current state is... | You can select... | Restrictions and notes |
|---|---|---|
| Online | Service Prep | All following conditions must first be met:
|
| Force Shutdown | Select Force Shutdown to confirm this change. Important: Once a shutdown operation is initiated, it cannot be canceled and IBM personnel is
required to return power to the cluster.
|
|
| Fence | Select Fence to manually fence the accessing cluster. You will get a panel for additional information before the fence action is initiated. For more information, refer to Fence or Unfence clusters. | |
| Service Pending | Force Service | You can select this option if you believe an operation has stalled and is
preventing the cluster from entering Service
Prep. Select Force Service to confirm this change.
Note: You can
place all but one cluster in a grid into Service mode but it is recommended that only one cluster is
in Service mode at a time. If more than one cluster is in Service mode, and you cancel Service mode
on one of them, then that cluster will not return to normal operation until Service mode is canceled
on all clusters in the grid or it is forced online.
|
| Return to Normal | You can select this option to cancel a previous Service Prep change and return
the cluster to the normal online
state. Select Return to Normal to confirm this change.
|
|
| Force Shutdown | Select Force Shutdown to confirm this change. Important: After a shutdown operation is initiated, it cannot be canceled, and IBM personnel
is required to return power to the cluster.
|
|
| Service | Return to Normal | Select this option to return the cluster to Normal mode. |
| Shutdown | Select this option to shut down the cluster. Important: Once a shutdown operation is initiated, it cannot be canceled.
|
|
| Shutdown (offline) | User interface not available | Once an offline cluster is powered on it attempts to return to normal. If no other clusters in the grid are available, you can skip hot token reconciliation. |
| Online-Pending or Shutdown Pending | Menu disabled | No options to change state are available when a cluster is in a pending state. |
- Hot tokens
-
A cluster in a grid configuration attempts to merge its token information with all the other clusters in the grid as it goes online. When no other clusters are available for this merge operation the cluster attempting to go online remains in the "going online", or blocked, state indefinitely as it waits for the other clusters to become available for the merge operation. If a pending merge operation is preventing the cluster from coming online, you are given the option to skip the merge step. Click the Skip Step button to skip the merge operation. This button is only available if the cluster is in a blocked state waiting to share pending updates with one or more unavailable clusters.
If you click the Skip Step button, pending updates against the local cluster can remain undetected until the unavailable cluster(s) become available.
- Ownership takeover
-
If ownership takeover was set at any of the peers, the possibility exists that old data can surface to the host if the cluster is forced online. Therefore, before attempting to force this cluster online it is important to know whether any peer clusters have ever enabled ownership takeover mode against this cluster while it was unavailable. In addition, if this cluster is currently in service, automatic ownership takeover from unavailable peers is also likely and should be considered before attempting to force this cluster online. If multiple clusters have been offline and must be forced back online, force them back online in the reverse order that they went down in (for example, the last cluster down is the first cluster up). This assures that the most current cluster is available first to educate the rest of the clusters forced online.
If Autonomic Ownership Takeover Manager (AOTM) is installed and configured, it will attempt to determine if all unavailable peer clusters are actually in a failed state. If it determines that the unavailable cluster is not in a failed state, it will block an attempt to force the cluster online. If the unavailable cluster is not actually in a failed state, the forced online cluster could be taking ownership of volumes that it should not be. If AOTM discovers that all unavailable peers have failed and network issues are not to blame, this cluster will then be forced into an online state. Once online, AOTM may further allow ownership takeover against the unavailable clusters if the AOTM option is enabled. Additionally, manual ownership takeover can be enabled if necessary.
- Shutdown restrictions
-
You can shut down only the cluster that you are logged in to. To shut down another cluster, you must log out of the current cluster and log in to the cluster you want to shut down.
Before you shut down the TS7700, you must decide whether your circumstances provide adequate time to complete a clean shutdown. A clean shutdown is not required, but is recommended for a TS7700 Grid configuration. A clean shutdown requires you to first place the cluster in service mode to ensure that no jobs are being processed during a shutdown operation. If you cannot place the cluster in service mode, you can force a shutdown of the cluster.
CAUTION:A forced shutdown can result in lost access to data and job failure. It should be used only when immediate shutdown with power off is required.